Line | Branch | Exec | Source |
---|---|---|---|
1 | /* | ||
2 | * DSP Group TrueSpeech compatible decoder | ||
3 | * Copyright (c) 2005 Konstantin Shishkov | ||
4 | * | ||
5 | * This file is part of FFmpeg. | ||
6 | * | ||
7 | * FFmpeg is free software; you can redistribute it and/or | ||
8 | * modify it under the terms of the GNU Lesser General Public | ||
9 | * License as published by the Free Software Foundation; either | ||
10 | * version 2.1 of the License, or (at your option) any later version. | ||
11 | * | ||
12 | * FFmpeg is distributed in the hope that it will be useful, | ||
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | ||
15 | * Lesser General Public License for more details. | ||
16 | * | ||
17 | * You should have received a copy of the GNU Lesser General Public | ||
18 | * License along with FFmpeg; if not, write to the Free Software | ||
19 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | ||
20 | */ | ||
21 | |||
22 | #include "libavutil/channel_layout.h" | ||
23 | #include "libavutil/mem_internal.h" | ||
24 | |||
25 | #include "avcodec.h" | ||
26 | #include "bswapdsp.h" | ||
27 | #include "codec_internal.h" | ||
28 | #include "decode.h" | ||
29 | #include "get_bits.h" | ||
30 | |||
31 | #include "truespeech_data.h" | ||
32 | /** | ||
33 | * @file | ||
34 | * TrueSpeech decoder. | ||
35 | */ | ||
36 | |||
37 | /** | ||
38 | * TrueSpeech decoder context | ||
39 | */ | ||
40 | typedef struct TSContext { | ||
41 | BswapDSPContext bdsp; | ||
42 | /* input data */ | ||
43 | DECLARE_ALIGNED(16, uint8_t, buffer)[32]; | ||
44 | int16_t vector[8]; ///< input vector: 5/5/4/4/4/3/3/3 | ||
45 | int offset1[2]; ///< 8-bit value, used in one copying offset | ||
46 | int offset2[4]; ///< 7-bit value, encodes offsets for copying and for two-point filter | ||
47 | int pulseoff[4]; ///< 4-bit offset of pulse values block | ||
48 | int pulsepos[4]; ///< 27-bit variable, encodes 7 pulse positions | ||
49 | int pulseval[4]; ///< 7x2-bit pulse values | ||
50 | int flag; ///< 1-bit flag, shows how to choose filters | ||
51 | /* temporary data */ | ||
52 | int filtbuf[146]; // some big vector used for storing filters | ||
53 | int prevfilt[8]; // filter from previous frame | ||
54 | int16_t tmp1[8]; // coefficients for adding to out | ||
55 | int16_t tmp2[8]; // coefficients for adding to out | ||
56 | int16_t tmp3[8]; // coefficients for adding to out | ||
57 | int16_t cvector[8]; // correlated input vector | ||
58 | int filtval; // gain value for one function | ||
59 | int16_t newvec[60]; // tmp vector | ||
60 | int16_t filters[32]; // filters for every subframe | ||
61 | } TSContext; | ||
62 | |||
63 | 2 | static av_cold int truespeech_decode_init(AVCodecContext * avctx) | |
64 | { | ||
65 | 2 | TSContext *c = avctx->priv_data; | |
66 | |||
67 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
|
2 | if (avctx->ch_layout.nb_channels != 1) { |
68 | ✗ | avpriv_request_sample(avctx, "Channel count %d", avctx->ch_layout.nb_channels); | |
69 | ✗ | return AVERROR_PATCHWELCOME; | |
70 | } | ||
71 | |||
72 | 2 | av_channel_layout_uninit(&avctx->ch_layout); | |
73 | 2 | avctx->ch_layout = (AVChannelLayout)AV_CHANNEL_LAYOUT_MONO; | |
74 | 2 | avctx->sample_fmt = AV_SAMPLE_FMT_S16; | |
75 | |||
76 | 2 | ff_bswapdsp_init(&c->bdsp); | |
77 | |||
78 | 2 | return 0; | |
79 | } | ||
80 | |||
81 | 1442 | static void truespeech_read_frame(TSContext *dec, const uint8_t *input) | |
82 | { | ||
83 | GetBitContext gb; | ||
84 | |||
85 | 1442 | dec->bdsp.bswap_buf((uint32_t *) dec->buffer, (const uint32_t *) input, 8); | |
86 | 1442 | init_get_bits(&gb, dec->buffer, 32 * 8); | |
87 | |||
88 | 1442 | dec->vector[7] = ts_codebook[7][get_bits(&gb, 3)]; | |
89 | 1442 | dec->vector[6] = ts_codebook[6][get_bits(&gb, 3)]; | |
90 | 1442 | dec->vector[5] = ts_codebook[5][get_bits(&gb, 3)]; | |
91 | 1442 | dec->vector[4] = ts_codebook[4][get_bits(&gb, 4)]; | |
92 | 1442 | dec->vector[3] = ts_codebook[3][get_bits(&gb, 4)]; | |
93 | 1442 | dec->vector[2] = ts_codebook[2][get_bits(&gb, 4)]; | |
94 | 1442 | dec->vector[1] = ts_codebook[1][get_bits(&gb, 5)]; | |
95 | 1442 | dec->vector[0] = ts_codebook[0][get_bits(&gb, 5)]; | |
96 | 1442 | dec->flag = get_bits1(&gb); | |
97 | |||
98 | 1442 | dec->offset1[0] = get_bits(&gb, 4) << 4; | |
99 | 1442 | dec->offset2[3] = get_bits(&gb, 7); | |
100 | 1442 | dec->offset2[2] = get_bits(&gb, 7); | |
101 | 1442 | dec->offset2[1] = get_bits(&gb, 7); | |
102 | 1442 | dec->offset2[0] = get_bits(&gb, 7); | |
103 | |||
104 | 1442 | dec->offset1[1] = get_bits(&gb, 4); | |
105 | 1442 | dec->pulseval[1] = get_bits(&gb, 14); | |
106 | 1442 | dec->pulseval[0] = get_bits(&gb, 14); | |
107 | |||
108 | 1442 | dec->offset1[1] |= get_bits(&gb, 4) << 4; | |
109 | 1442 | dec->pulseval[3] = get_bits(&gb, 14); | |
110 | 1442 | dec->pulseval[2] = get_bits(&gb, 14); | |
111 | |||
112 | 1442 | dec->offset1[0] |= get_bits1(&gb); | |
113 | 1442 | dec->pulsepos[0] = get_bits_long(&gb, 27); | |
114 | 1442 | dec->pulseoff[0] = get_bits(&gb, 4); | |
115 | |||
116 | 1442 | dec->offset1[0] |= get_bits1(&gb) << 1; | |
117 | 1442 | dec->pulsepos[1] = get_bits_long(&gb, 27); | |
118 | 1442 | dec->pulseoff[1] = get_bits(&gb, 4); | |
119 | |||
120 | 1442 | dec->offset1[0] |= get_bits1(&gb) << 2; | |
121 | 1442 | dec->pulsepos[2] = get_bits_long(&gb, 27); | |
122 | 1442 | dec->pulseoff[2] = get_bits(&gb, 4); | |
123 | |||
124 | 1442 | dec->offset1[0] |= get_bits1(&gb) << 3; | |
125 | 1442 | dec->pulsepos[3] = get_bits_long(&gb, 27); | |
126 | 1442 | dec->pulseoff[3] = get_bits(&gb, 4); | |
127 | 1442 | } | |
128 | |||
129 | 1442 | static void truespeech_correlate_filter(TSContext *dec) | |
130 | { | ||
131 | int16_t tmp[8]; | ||
132 | int i, j; | ||
133 | |||
134 |
2/2✓ Branch 0 taken 11536 times.
✓ Branch 1 taken 1442 times.
|
12978 | for(i = 0; i < 8; i++){ |
135 |
2/2✓ Branch 0 taken 10094 times.
✓ Branch 1 taken 1442 times.
|
11536 | if(i > 0){ |
136 | 10094 | memcpy(tmp, dec->cvector, i * sizeof(*tmp)); | |
137 |
2/2✓ Branch 0 taken 40376 times.
✓ Branch 1 taken 10094 times.
|
50470 | for(j = 0; j < i; j++) |
138 | 40376 | dec->cvector[j] += (tmp[i - j - 1] * dec->vector[i] + 0x4000) >> 15; | |
139 | } | ||
140 | 11536 | dec->cvector[i] = (8 - dec->vector[i]) >> 3; | |
141 | } | ||
142 |
2/2✓ Branch 0 taken 11536 times.
✓ Branch 1 taken 1442 times.
|
12978 | for(i = 0; i < 8; i++) |
143 | 11536 | dec->cvector[i] = (dec->cvector[i] * ts_decay_994_1000[i]) >> 15; | |
144 | |||
145 | 1442 | dec->filtval = dec->vector[0]; | |
146 | 1442 | } | |
147 | |||
148 | 1442 | static void truespeech_filters_merge(TSContext *dec) | |
149 | { | ||
150 | int i; | ||
151 | |||
152 |
2/2✓ Branch 0 taken 11 times.
✓ Branch 1 taken 1431 times.
|
1442 | if(!dec->flag){ |
153 |
2/2✓ Branch 0 taken 88 times.
✓ Branch 1 taken 11 times.
|
99 | for(i = 0; i < 8; i++){ |
154 | 88 | dec->filters[i + 0] = dec->prevfilt[i]; | |
155 | 88 | dec->filters[i + 8] = dec->prevfilt[i]; | |
156 | } | ||
157 | }else{ | ||
158 |
2/2✓ Branch 0 taken 11448 times.
✓ Branch 1 taken 1431 times.
|
12879 | for(i = 0; i < 8; i++){ |
159 | 11448 | dec->filters[i + 0]=(dec->cvector[i] * 21846 + dec->prevfilt[i] * 10923 + 16384) >> 15; | |
160 | 11448 | dec->filters[i + 8]=(dec->cvector[i] * 10923 + dec->prevfilt[i] * 21846 + 16384) >> 15; | |
161 | } | ||
162 | } | ||
163 |
2/2✓ Branch 0 taken 11536 times.
✓ Branch 1 taken 1442 times.
|
12978 | for(i = 0; i < 8; i++){ |
164 | 11536 | dec->filters[i + 16] = dec->cvector[i]; | |
165 | 11536 | dec->filters[i + 24] = dec->cvector[i]; | |
166 | } | ||
167 | 1442 | } | |
168 | |||
169 | 5768 | static void truespeech_apply_twopoint_filter(TSContext *dec, int quart) | |
170 | { | ||
171 | int16_t tmp[146 + 60], *ptr0, *ptr1; | ||
172 | const int16_t *filter; | ||
173 | int i, t, off; | ||
174 | |||
175 | 5768 | t = dec->offset2[quart]; | |
176 |
2/2✓ Branch 0 taken 358 times.
✓ Branch 1 taken 5410 times.
|
5768 | if(t == 127){ |
177 | 358 | memset(dec->newvec, 0, 60 * sizeof(*dec->newvec)); | |
178 | 358 | return; | |
179 | } | ||
180 |
2/2✓ Branch 0 taken 789860 times.
✓ Branch 1 taken 5410 times.
|
795270 | for(i = 0; i < 146; i++) |
181 | 789860 | tmp[i] = dec->filtbuf[i]; | |
182 | 5410 | off = (t / 25) + dec->offset1[quart >> 1] + 18; | |
183 | 5410 | off = av_clip(off, 0, 145); | |
184 | 5410 | ptr0 = tmp + 145 - off; | |
185 | 5410 | ptr1 = tmp + 146; | |
186 | 5410 | filter = ts_order2_coeffs + (t % 25) * 2; | |
187 |
2/2✓ Branch 0 taken 324600 times.
✓ Branch 1 taken 5410 times.
|
330010 | for(i = 0; i < 60; i++){ |
188 | 324600 | t = (ptr0[0] * filter[0] + ptr0[1] * filter[1] + 0x2000) >> 14; | |
189 | 324600 | ptr0++; | |
190 | 324600 | dec->newvec[i] = t; | |
191 | 324600 | ptr1[i] = t; | |
192 | } | ||
193 | } | ||
194 | |||
195 | 5768 | static void truespeech_place_pulses(TSContext *dec, int16_t *out, int quart) | |
196 | { | ||
197 | int16_t tmp[7]; | ||
198 | int i, j, t; | ||
199 | const int16_t *ptr1; | ||
200 | int16_t *ptr2; | ||
201 | int coef; | ||
202 | |||
203 | 5768 | memset(out, 0, 60 * sizeof(*out)); | |
204 |
2/2✓ Branch 0 taken 40376 times.
✓ Branch 1 taken 5768 times.
|
46144 | for(i = 0; i < 7; i++) { |
205 | 40376 | t = dec->pulseval[quart] & 3; | |
206 | 40376 | dec->pulseval[quart] >>= 2; | |
207 | 40376 | tmp[6 - i] = ts_pulse_scales[dec->pulseoff[quart] * 4 + t]; | |
208 | } | ||
209 | |||
210 | 5768 | coef = dec->pulsepos[quart] >> 15; | |
211 | 5768 | ptr1 = ts_pulse_values + 30; | |
212 | 5768 | ptr2 = tmp; | |
213 |
4/4✓ Branch 0 taken 135547 times.
✓ Branch 1 taken 591 times.
✓ Branch 2 taken 130370 times.
✓ Branch 3 taken 5177 times.
|
136138 | for(i = 0, j = 3; (i < 30) && (j > 0); i++){ |
214 | 130370 | t = *ptr1++; | |
215 |
2/2✓ Branch 0 taken 113066 times.
✓ Branch 1 taken 17304 times.
|
130370 | if(coef >= t) |
216 | 113066 | coef -= t; | |
217 | else{ | ||
218 | 17304 | out[i] = *ptr2++; | |
219 | 17304 | ptr1 += 30; | |
220 | 17304 | j--; | |
221 | } | ||
222 | } | ||
223 | 5768 | coef = dec->pulsepos[quart] & 0x7FFF; | |
224 | 5768 | ptr1 = ts_pulse_values; | |
225 |
4/4✓ Branch 0 taken 141217 times.
✓ Branch 1 taken 180 times.
✓ Branch 2 taken 135629 times.
✓ Branch 3 taken 5588 times.
|
141397 | for(i = 30, j = 4; (i < 60) && (j > 0); i++){ |
226 | 135629 | t = *ptr1++; | |
227 |
2/2✓ Branch 0 taken 112557 times.
✓ Branch 1 taken 23072 times.
|
135629 | if(coef >= t) |
228 | 112557 | coef -= t; | |
229 | else{ | ||
230 | 23072 | out[i] = *ptr2++; | |
231 | 23072 | ptr1 += 30; | |
232 | 23072 | j--; | |
233 | } | ||
234 | } | ||
235 | |||
236 | 5768 | } | |
237 | |||
238 | 5768 | static void truespeech_update_filters(TSContext *dec, int16_t *out, int quart) | |
239 | { | ||
240 | int i; | ||
241 | |||
242 | 5768 | memmove(dec->filtbuf, &dec->filtbuf[60], 86 * sizeof(*dec->filtbuf)); | |
243 |
2/2✓ Branch 0 taken 346080 times.
✓ Branch 1 taken 5768 times.
|
351848 | for(i = 0; i < 60; i++){ |
244 | 346080 | dec->filtbuf[i + 86] = out[i] + dec->newvec[i] - (dec->newvec[i] >> 3); | |
245 | 346080 | out[i] += dec->newvec[i]; | |
246 | } | ||
247 | 5768 | } | |
248 | |||
249 | 5768 | static void truespeech_synth(TSContext *dec, int16_t *out, int quart) | |
250 | { | ||
251 | int i,k; | ||
252 | int t[8]; | ||
253 | int16_t *ptr0, *ptr1; | ||
254 | |||
255 | 5768 | ptr0 = dec->tmp1; | |
256 | 5768 | ptr1 = dec->filters + quart * 8; | |
257 |
2/2✓ Branch 0 taken 346080 times.
✓ Branch 1 taken 5768 times.
|
351848 | for(i = 0; i < 60; i++){ |
258 | 346080 | int sum = 0; | |
259 |
2/2✓ Branch 0 taken 2768640 times.
✓ Branch 1 taken 346080 times.
|
3114720 | for(k = 0; k < 8; k++) |
260 | 2768640 | sum += ptr0[k] * (unsigned)ptr1[k]; | |
261 | 346080 | sum = out[i] + ((int)(sum + 0x800U) >> 12); | |
262 | 346080 | out[i] = av_clip(sum, -0x7FFE, 0x7FFE); | |
263 |
2/2✓ Branch 0 taken 2422560 times.
✓ Branch 1 taken 346080 times.
|
2768640 | for(k = 7; k > 0; k--) |
264 | 2422560 | ptr0[k] = ptr0[k - 1]; | |
265 | 346080 | ptr0[0] = out[i]; | |
266 | } | ||
267 | |||
268 |
2/2✓ Branch 0 taken 46144 times.
✓ Branch 1 taken 5768 times.
|
51912 | for(i = 0; i < 8; i++) |
269 | 46144 | t[i] = (ts_decay_35_64[i] * ptr1[i]) >> 15; | |
270 | |||
271 | 5768 | ptr0 = dec->tmp2; | |
272 |
2/2✓ Branch 0 taken 346080 times.
✓ Branch 1 taken 5768 times.
|
351848 | for(i = 0; i < 60; i++){ |
273 | 346080 | int sum = 0; | |
274 |
2/2✓ Branch 0 taken 2768640 times.
✓ Branch 1 taken 346080 times.
|
3114720 | for(k = 0; k < 8; k++) |
275 | 2768640 | sum += ptr0[k] * t[k]; | |
276 |
2/2✓ Branch 0 taken 2422560 times.
✓ Branch 1 taken 346080 times.
|
2768640 | for(k = 7; k > 0; k--) |
277 | 2422560 | ptr0[k] = ptr0[k - 1]; | |
278 | 346080 | ptr0[0] = out[i]; | |
279 | 346080 | out[i] += (- sum) >> 12; | |
280 | } | ||
281 | |||
282 |
2/2✓ Branch 0 taken 46144 times.
✓ Branch 1 taken 5768 times.
|
51912 | for(i = 0; i < 8; i++) |
283 | 46144 | t[i] = (ts_decay_3_4[i] * ptr1[i]) >> 15; | |
284 | |||
285 | 5768 | ptr0 = dec->tmp3; | |
286 |
2/2✓ Branch 0 taken 346080 times.
✓ Branch 1 taken 5768 times.
|
351848 | for(i = 0; i < 60; i++){ |
287 | 346080 | int sum = out[i] * (1 << 12); | |
288 |
2/2✓ Branch 0 taken 2768640 times.
✓ Branch 1 taken 346080 times.
|
3114720 | for(k = 0; k < 8; k++) |
289 | 2768640 | sum += ptr0[k] * t[k]; | |
290 |
2/2✓ Branch 0 taken 2422560 times.
✓ Branch 1 taken 346080 times.
|
2768640 | for(k = 7; k > 0; k--) |
291 | 2422560 | ptr0[k] = ptr0[k - 1]; | |
292 | 346080 | ptr0[0] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE); | |
293 | |||
294 | 346080 | sum = ((ptr0[1] * (dec->filtval - (dec->filtval >> 2))) >> 4) + sum; | |
295 | 346080 | sum = sum - (sum >> 3); | |
296 | 346080 | out[i] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE); | |
297 | } | ||
298 | 5768 | } | |
299 | |||
300 | 1442 | static void truespeech_save_prevvec(TSContext *c) | |
301 | { | ||
302 | int i; | ||
303 | |||
304 |
2/2✓ Branch 0 taken 11536 times.
✓ Branch 1 taken 1442 times.
|
12978 | for(i = 0; i < 8; i++) |
305 | 11536 | c->prevfilt[i] = c->cvector[i]; | |
306 | 1442 | } | |
307 | |||
308 | 721 | static int truespeech_decode_frame(AVCodecContext *avctx, AVFrame *frame, | |
309 | int *got_frame_ptr, AVPacket *avpkt) | ||
310 | { | ||
311 | 721 | const uint8_t *buf = avpkt->data; | |
312 | 721 | int buf_size = avpkt->size; | |
313 | 721 | TSContext *c = avctx->priv_data; | |
314 | |||
315 | int i, j; | ||
316 | int16_t *samples; | ||
317 | int iterations, ret; | ||
318 | |||
319 | 721 | iterations = buf_size / 32; | |
320 | |||
321 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 721 times.
|
721 | if (!iterations) { |
322 | ✗ | av_log(avctx, AV_LOG_ERROR, | |
323 | "Too small input buffer (%d bytes), need at least 32 bytes\n", buf_size); | ||
324 | ✗ | return -1; | |
325 | } | ||
326 | |||
327 | /* get output buffer */ | ||
328 | 721 | frame->nb_samples = iterations * 240; | |
329 |
1/2✗ Branch 1 not taken.
✓ Branch 2 taken 721 times.
|
721 | if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) |
330 | ✗ | return ret; | |
331 | 721 | samples = (int16_t *)frame->data[0]; | |
332 | |||
333 | 721 | memset(samples, 0, iterations * 240 * sizeof(*samples)); | |
334 | |||
335 |
2/2✓ Branch 0 taken 1442 times.
✓ Branch 1 taken 721 times.
|
2163 | for(j = 0; j < iterations; j++) { |
336 | 1442 | truespeech_read_frame(c, buf); | |
337 | 1442 | buf += 32; | |
338 | |||
339 | 1442 | truespeech_correlate_filter(c); | |
340 | 1442 | truespeech_filters_merge(c); | |
341 | |||
342 |
2/2✓ Branch 0 taken 5768 times.
✓ Branch 1 taken 1442 times.
|
7210 | for(i = 0; i < 4; i++) { |
343 | 5768 | truespeech_apply_twopoint_filter(c, i); | |
344 | 5768 | truespeech_place_pulses (c, samples, i); | |
345 | 5768 | truespeech_update_filters(c, samples, i); | |
346 | 5768 | truespeech_synth (c, samples, i); | |
347 | 5768 | samples += 60; | |
348 | } | ||
349 | |||
350 | 1442 | truespeech_save_prevvec(c); | |
351 | } | ||
352 | |||
353 | 721 | *got_frame_ptr = 1; | |
354 | |||
355 | 721 | return buf_size; | |
356 | } | ||
357 | |||
358 | const FFCodec ff_truespeech_decoder = { | ||
359 | .p.name = "truespeech", | ||
360 | CODEC_LONG_NAME("DSP Group TrueSpeech"), | ||
361 | .p.type = AVMEDIA_TYPE_AUDIO, | ||
362 | .p.id = AV_CODEC_ID_TRUESPEECH, | ||
363 | .priv_data_size = sizeof(TSContext), | ||
364 | .init = truespeech_decode_init, | ||
365 | FF_CODEC_DECODE_CB(truespeech_decode_frame), | ||
366 | .p.capabilities = AV_CODEC_CAP_DR1, | ||
367 | }; | ||
368 |